Hardik
Sankhla
— Building Systems That Think
I design and build autonomous ML infrastructure — from LLM pipelines to agent architectures. Not wrappers. Not demos. Production systems that scale.
3+Years
12+Systems
∞Iterations
— Building Systems That Think
I design and build autonomous ML infrastructure — from LLM pipelines to agent architectures. Not wrappers. Not demos. Production systems that scale.
// 01 — About
I specialize in the unglamorous, high-leverage work: designing ML systems that run reliably at scale. Vector databases, retrieval pipelines, agent loops, evaluation frameworks.
My mental model is engineering-first. I build for correctness, observability, and longevity — not for demos.
// 02 — Domains
End-to-end ML pipelines: data ingestion, feature stores, model serving, monitoring. Built to run in production, not Jupyter.
Training pipelines, distributed compute, experiment tracking. Fine-tuning LLMs with LoRA. Scaling from prototype to cluster.
Multi-agent orchestration, tool use, memory systems, evaluation loops. Agents that actually complete tasks — not hallucinate them.
// 03 — Work
Context-aware support assistant that improved answer relevance and reduced repetitive support load with retrieval, memory, and safety guardrails.
Real-time quality inspection pipeline that improved defect detection consistency and accelerated root-cause analysis on production lines.
Evaluation and telemetry framework that made LLM releases safer through repeatable quality checks and production observability.
Personalized recommendation service that improved relevance and supported controlled experimentation with low-latency APIs.
// 04 — Identity
// 05 — Contact
Open to research collaborations, full-time roles, and consulting on ML infrastructure problems. If you're building something ambitious, reach out.